Talking to AI: Prompt Guide & Examples from the Kitchen

I’ve been teaching people how to use generative AI tools for going on three years now. From supporting individuals one-on-one to training large teams, I’ve seen that everyone starts with the same fundamental question: "What can this tool actually do for me?"

When it comes to GenAI, curiosity is your best friend. It is the key to uncovering what to talk to these tools about and finding those "lightbulb" moments where the technology actually makes your life easier. The real question is - what can't these tools help with? If you have a problem, scenario, want to learn more, etc - stop wondering how the tool can help and simply ask it for help.

Before we dive in, let's clear up some common terms you’ll hear:

• GenAI: Short for Generative Artificial Intelligence—the technology that creates new content (like text or images).
• Model: The specific version or "brain" of the AI being used (e.g., Gemini 1.5 Pro or GPT-4o). It defines the intelligence level and capabilities of the software.
• Prompt: The instruction or message you send to the model to tell it what to do.

The answer to how these tools can help is simpler than you might think. Any time you have a thought that starts with"I wonder..." or "How can I...", you have a perfect candidate for an AI prompt. But a gap that many people have is in the execution: how do you actually type out a message to get the output you are looking for?

A common assumption is that tools like Gemini or ChatGPT "know best." In reality, these tools are just waiting for you to tell them what to do and how you want it done. They aren't mind readers; they are highly capable assistants waiting for clear instructions.

Let’s chat about how to craft an effective message to a generative AI tool to get exactly what you want, using the kitchen as our training ground.

How to Build a Message: The Google Framework

Whenever I start training a new group, I always teach the prompt format Google outlines in their official guide as the first step: Persona, Task, Context, and Format. It gives you a consistent structure to practice again and again on your own.

To help you remember the building blocks, we use the acronym PTCF:

You don’t need to use every block every time. However, practicing this in the kitchen helps you see the "pieces of the puzzle" needed to tackle any other task with AI.

Back to the Kitchen: The Framework in Action

The best way to see how these building blocks work is to look at a real-world scenario. I recently used this exact framework to develop an authentic Pad Thai recipe.

1. The Persona: I told Gemini to act as a Thai street food expert.
2. The Task: I asked it to create a recipe that avoided Westernized shortcuts.
3. The Context: I explained that I had done some research, and because of mixed opinions online, it was up in the air if oyster sauce was a component I actually wanted to use.
4. The Format: I asked for the recipe to be delivered in a clear outline so I could follow it easily while cooking.

The Result: Because I provided these specific pieces, the AI explained that authentic Pad Thai relies on a balance of tamarind, fish sauce, and palm sugar—confirming that oyster sauce is a non-traditional shortcut.

The Format was the real win in the kitchen. By delivering a structured outline, the AI clearly separated the sauce prep from the stir-fry steps. This allowed me to see everything at a glance and eliminated the need to scroll up and down between ingredients and instructions while cooking.

Why the Details Matter

Context and instruction are essential for the model to complete your task in the way you want it to be completed. If you don’t provide the details, the model has to make assumptions about your intent.

The more information you can provide in the very first message, the more likely the model will be closer to an output you can use on the first try. That said, it is normal to iterate or engage in a bit of back-and-forth.

You are the "Human in the Loop"

By reviewing the information put out by a model, you are always playing the role of the Human in the Loop. You are the one who decides if the output meets your needs. You are the head chef; the AI is simply there to help you process information more efficiently.

Real-Life Scenarios: Give it a Try

I’m sure you can relate to these kitchen moments where you’re mid-recipe and hit a wall. The next time you run into one of these, use the PTCF format to get help from your GenAI tool:

• The "Is it Done?" Check: You’re looking at a cake in the oven and don’t know how to tell if it’s actually baked through without sinking the middle.
• • Persona: Professional Pastry Chef
  • Task: Troubleshooting a bake
  • Context: 9-inch vanilla sponge cake, 30 mins in at 350°F, top is browning but the center looks wobbly.
  • Format: Simple bulleted checklist of signs to look for.
• The Meringue Mystery: You’ve been whisking for ages and are unsure if your meringue is whipped enough to hold a peak.
• • Persona: Culinary Instructor
  • Task: Visual guide for egg whites
  • Context: Hand-whisking 4 egg whites with sugar for a Pavlova; they look glossy but soft.
  • Format: Short descriptive paragraph.
• The Substitution Struggle: You’re missing an ingredient and need a substitute that won’t impact the final texture.
• • Persona: Food Scientist
  • Task: Ingredient substitution
  • Context: Out of buttermilk for a scone recipe; need a swap that maintains the acidity and rise.
  • Format: A list of 2-3 options with ratios.

A Simple Template to Start

The next time you’re stuck, try using this basic outline:

"Act as a [Persona]. I need you to [Task]. Here is the info I have: [Context]. Please give me the answer in [Format]."

LMK if you want a part two of this prompting discussion to learn some elevated tips when prompting genAI tools!